| 1. |
|
|
| 2. |
- Devarakota, Pandu Ranga Rao, 1978-
(författare)
-
Classification and Localization of Vehicle Occupants Using 3D Range Images
- 2008
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- This thesis deals with the problem of classifying automotive vehicle occupants and estimating their position. This information is critical in designing future smart airbag systems providing maximum protection for passengers. According to the American National Highway Traffic Safety Administration (NHTSA), since 1990, in the USA, 227 deaths have been attributed to airbags deployed in low-speed crashes which included 119 children, and 22 infants. In these cases, intelligent deployment of the airbag, based on the type and position of occupant could have avoided these fatalities. Current commercial classification systems based on traditional sensors, such as pressure sensors are not able to detect the position of occupants. Vision-based systems are advantageous over pressure sensor based systems, as they can provide additional functionalities like dynamic occupant position analysis or child seat orientation detection. On the other hand, vision-based systems have to cope with several challenges, such as, illumination conditions, temperature, humidity, large variation of scenes, cost, and computational aspects. This thesis presents new pattern recognition techniques for classifying, localizing and tracking vehicle occupants using a low-resolution 3-D optical time-of-flight range camera. This sensor is capable of providing directly a dense range image, independent of the illumination conditions and object textures. Based on this technology, IEE S.A. is presently developing a camera system for the application of occupant classification. A prototype of this camera has been the basis for this study. The first part of the thesis presents the problem of occupant classification. Herein, we investigate geometric feature extraction methods to discriminate between different occupant types. We develop features that are invariant under rotation and translation. A method for reducing the size of the feature set is analyzed with emphasis on robustness and low computational complexity while maintaining highly discriminative information. In addition, several classification methods are studied including Bayes quadratic classifier, Gaussian Mixture Model (GMM) classifier and polynomial classifier. We propose the use of a cluster based linear regression classifier using a polynomial kernel which is particularly well suited to coping with large variations within each class. Full scale experiments have been conducted which demonstrate that a classification reliability of almost 100\% can be achieved with the reduced feature set in combination with a cluster based classifier. In this safety critical application, it is equally important to address the problem of reliability estimation for the system. State-of-the-art methods to estimate the reliability of the classification are based either on classification output or based on density estimation. The second part of the thesis treats estimation of the reliability of the pattern classification system. Herein, a novel reliability measure is proposed for classification output which takes into account the local density of training data. Experiments verify that this reliability measure outperforms state-of-the-art methods in many cases. Lastly, the problem of dynamically detecting out-of-position occupants is addressed in the third part of the thesis. This task requires detecting and localizing the position of the occupant's head. Traditional head detection methods, such as detecting head-like objects in the image by analyzing the local shapes are not robust with the current sensor. Many regions in a scene such as the shoulder or the elbow of the occupant can be incorrectly detected as the head. In order to cope with these challenges, we exploit topology information in the range image. A modified Reeb graph technique has been developed that extracts a topological skeleton of the 3D contour that is invariant under rotation and translations. Results verify that the Reeb graph detects successfully the head i.e., the head always corresponds to one of the end points of the skeleton. Subsequently, a data association algorithm to select the correct head candidate out of the Reeb graph candidates is presented. Results show that the resulting head detection algorithm based on Reeb graphs is robust under scene changes.
|
|
| 3. |
- Ghanem, Sally, et al.
(författare)
-
Robust Group Subspace Recovery: A New Approach for Multi-Modality Data Fusion
- 2020
-
Ingår i: IEEE Sensors Journal. - 1558-1748 .- 1530-437X. ; 20:20, s. 12307-12316
-
Tidskriftsartikel (refereegranskat)abstract
- Robust Subspace Recovery (RoSuRe) algorithm was recently introduced as a principled and numerically efficient algorithm that unfolds underlying Unions of Subspaces (UoS) structure, present in the data. The union of Subspaces (UoS) is capable of identifying more complex trends in data sets than simple linear models. We build on and extend RoSuRe to prospect the structure of different data modalities individually. We propose a novel multi-modal data fusion approach based on group sparsity which we refer to as Robust Group Subspace Recovery (RoGSuRe). Relying on a bi-sparsity pursuit paradigm and non-smooth optimization techniques, the introduced framework learns a new joint representation of the time series from different data modalities, respecting an underlying UoS model. We subsequently integrate the obtained structures to form a unified subspace structure. The proposed approach exploits the structural dependencies between the different modalities data to cluster the associated target objects. The resulting fusion of the unlabeled sensors' data from experiments on audio and magnetic data has shown that our method is competitive with other state of the art subspace clustering methods. The resulting UoS structure is employed to classify newly observed data points, highlighting the abstraction capacity of the proposed method.
|
|
| 4. |
- Huang, Yuming, et al.
(författare)
-
Community Detection and Improved Detectability in Multiplex Networks
- 2020
-
Ingår i: IEEE Transactions on Network Science and Engineering. - 2327-4697. ; 7:3, s. 1697-1709
-
Tidskriftsartikel (refereegranskat)abstract
- Belief propagation is a technique to optimize probabilistic graphical models, and has been used to solve the community detection problem for networks described by the stochastic block model. In this work, we investigate the community detection problem in multiplex networks with generic community label constraints using the belief propagation algorithm. Our main contribution is a generative model that does not assume consistent communities between layers and allows a potentially heterogeneous community structure, suitable in many real world multiplex networks, such as social networks. We show by numerical experiments that in the presence of consistent communities between different layers, consistent communities are matched, and the detectability is improved over single layers. We compare it with a "correlated model" which has the prior knowledge of community correlation between layers. Similar detectability improvement is obtained, even though our model has much milder assumptions than the "correlated model". When the network has heterogeneous community structures, our model is shown to yield a better detection performance over a certain parameter range.
|
|
| 5. |
- Huang, Yuming, et al.
(författare)
-
Fusion of Community Structures in Multiplex Networks by Label Constraints
- 2018
-
Ingår i: 26th European Signal Processing Conference (EUSIPCO). ; , s. 887-891
-
Konferensbidrag (refereegranskat)abstract
- We develop a Belief Propagation algorithm for community detection problem in multiplex networks, which more accurately represents many real-world systems. Previous works have established that real world multiplex networks exhibit redundant structures/communities, and that community detection performance improves by aggregating (fusing) redundant layers which are generated from the same Stochastic Block Model (SBM). We introduce a probability model for generic multiplex networks, aiming to fuse community structure across layers, without assuming or seeking the same SBM generative model for different layers. Numerical experiment shows that our model finds out consistent communities between layers and yields a significant detectability improvement over the single layer architecture. Our model also achieves a comparable performance to a reference model where we assume consistent communities in prior. Finally we compare our method with multilayer modularity optimization in heterogeneous networks, and show that our method detects correct community labels more reliably.
|
|
| 6. |
- Jiang, Bo, et al.
(författare)
-
Dynamic graph learning: A structure-driven approach
- 2021
-
Ingår i: Mathematics. - : MDPI AG. - 2227-7390. ; 9:2, s. 1-20
-
Tidskriftsartikel (refereegranskat)abstract
- The purpose of this paper is to infer a dynamic graph as a global (collective) model of time-varying measurements at a set of network nodes. This model captures both pairwise as well as higher order interactions (i.e., more than two nodes) among the nodes. The motivation of this work lies in the search for a connectome model which properly captures brain functionality across all regions of the brain, and possibly at individual neurons. We formulate it as an optimization problem, a quadratic objective functional and tensor information of observed node signals over short time intervals. The proper regularization constraints reflect the graph smoothness and other dynamics involving the underlying graph’s Laplacian, as well as the time evolution smoothness of the underlying graph. The resulting joint optimization is solved by a continuous relaxation of the weight parameters and an introduced novel gradient-projection scheme. While the work may be applicable to any time-evolving data set (e.g., fMRI), we apply our algorithm to a real-world dataset comprising recorded activities of individual brain cells. The resulting model is shown to be not only viable but also efficiently computable.
|
|
| 7. |
- Mahdizadehaghdam, Shahin, et al.
(författare)
-
Deep Dictionary Learning: A PARametric NETwork Approach
- 2019
-
Ingår i: IEEE Transactions on Image Processing. - 1941-0042 .- 1057-7149. ; 28:10, s. 4790-4802
-
Tidskriftsartikel (refereegranskat)abstract
- Deep dictionary learning seeks multiple dictionaries at different image scales to capture complementary coherent characteristics. We propose a method for learning a hierarchy of synthesis dictionaries with an image classification goal. The dictionaries and classification parameters are trained by a classification objective, and the sparse features are extracted by reducing a reconstruction loss in each layer. The reconstruction objectives in some sense regularize the classification problem and inject source signal information in the extracted features. The performance of the proposed hierarchical method increases by adding more layers, which consequently makes this model easier to tune and adapt. The proposed algorithm furthermore shows a remarkably lower fooling rate in the presence of adversarial perturbation. The validation of the proposed approach is based on its classification performance using four benchmark datasets and is compared to a Convolutional Neural Network (CNN) of similar size.
|
|
| 8. |
- Panahi, Ashkan, 1986, et al.
(författare)
-
Demystifying Deep Learning: a Geometric Approach to Iterative Projections
- 2018
-
Ingår i: ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings. - 1520-6149.
-
Konferensbidrag (refereegranskat)abstract
- Parametric approaches to Learning, such as deep learning (DL), are highly popular in nonlinear regression, in spite of their extremely difficult training with their increasing complexity (e.g. number of layers in DL). In this paper, we present an alternative semi-parametric framework which foregoes the ordinarily required feedback, by introducing the novel idea of geometric regularization. We show that certain deep learning techniques such as residual network (ResNet) architecture are closely related to our approach. Hence, our technique can be used to analyze these types of deep learning. Moreover, we present preliminary results which confirm that our approach can be easily trained to obtain complex structures.
|
|
| 9. |
- Panahi, Ashkan, 1986, et al.
(författare)
-
Robust Subspace Clustering by Bi-Sparsity Pursuit: Guarantees and Sequential Algorithm
- 2018
-
Ingår i: IEEE Winter Conference on Applications of Computer Vision. - 9781538648865 ; , s. 1302-1311
-
Konferensbidrag (refereegranskat)abstract
- We consider subspace clustering under sparse noise, for which a non-convex optimization framework based on sparse data representations has been recently developed. This setup is suitable for a large variety of applications with high dimensional data, such as image processing, which is naturally decomposed into a sparse unstructured foreground and a background residing in a union of low-dimensional subspaces. In this framework, we further discuss both performance and implementation of the key optimization problem. We provide an analysis of this optimization problem demonstrating that our approach is capable of recovering linear subspaces as a local optimal solution for sufficiently large data sets and sparse noise vectors. We also propose a sequential algorithmic solution, which is particularly useful for extremely large data sets and online vision applications such as video processing.
|
|
| 10. |
- Rahbar, Arman, 1992, et al.
(författare)
-
Recovery Bounds on Class-Based Optimal Transport: A Sum-of-Norms Regularization Framework
- 2023
-
Ingår i: Proceedings of the 38th Conference on Uncertainty in Artificial Intelligence, UAI 2022. - 2640-3498. ; 202, s. 28549-28577
-
Konferensbidrag (refereegranskat)abstract
- We develop a novel theoretical framework for understating Optimal Transport (OT) schemes respecting a class structure. For this purpose, we propose a convex OT program with a sum-of-norms regularization term, which provably recovers the underlying class structure under geometric assumptions. Furthermore, we derive an accelerated proximal algorithm with a closed-form projection and proximal operator scheme, thereby affording a more scalable algorithm for computing optimal transport plans. We provide a novel argument for the uniqueness of the optimum even in the absence of strong convexity. Our experiments show that the new regularizer not only results in a better preservation of the class structure in the data but also yields additional robustness to the data geometry, compared to previous regularizers.
|
|
